Re: [PATCH v3 1/7] kernel/fork: convert vma assignment to a memcpy

[Suren Baghdasaryan]

On Thu, Jan 26, 2023 at 3:52 AM Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Jan 25, 2023 at 05:34:49PM -0800, Andrew Morton wrote:
> > On Wed, 25 Jan 2023 16:50:01 -0800 Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
> >
> > > On Wed, Jan 25, 2023 at 4:22 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Wed, 25 Jan 2023 15:35:48 -0800 Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
> > > >
> > > > > Convert vma assignment in vm_area_dup() to a memcpy() to prevent compiler
> > > > > errors when we add a const modifier to vma->vm_flags.
> > > > >
> > > > > ...
> > > > >
> > > > > --- a/kernel/fork.c
> > > > > +++ b/kernel/fork.c
> > > > > @@ -482,7 +482,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
> > > > >                * orig->shared.rb may be modified concurrently, but the clone
> > > > >                * will be reinitialized.
> > > > >                */
> > > > > -             *new = data_race(*orig);
> > > > > +             memcpy(new, orig, sizeof(*new));
> > > >
> > > > The data_race() removal is unchangelogged?
> > >
> > > True. I'll add a note in the changelog about that. Ideally I would
> > > like to preserve it but I could not find a way to do that.
> > >
> >
> > Perhaps Paul can comment?
> >
> > I wonder if KCSAN knows how to detect this race, given that it's now in
> > a memcpy.  I assume so.
>
> data_race() is just wrapping an expression around
> __kcsan_[en|dis]able_current and ensuring the expression is evaluated once
> and returning the correct type. I believe the following should be sufficient.

Thanks for the suggestion, Mel! I'll try that.

>
> diff --git a/kernel/fork.c b/kernel/fork.c
> index 9f7fe3541897..1b30ee568e02 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -472,7 +472,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
>                  * orig->shared.rb may be modified concurrently, but the clone
>                  * will be reinitialized.
>                  */
> -               *new = data_race(*orig);
> +               data_race(memcpy(new, orig, sizeof(*new)));
>                 INIT_LIST_HEAD(&new->anon_vma_chain);
>                 dup_anon_vma_name(orig, new);
>         }
>
> I don't see how memcpy could automagically figure out whether the memcpy
> is prone to races or not in an arbitrary context.
>
> Assuming using data_race this way is ok then
>
> Acked-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>

Thanks!

>
> --
> Mel Gorman
> SUSE Labs